A Parallel Computing Dynamic Task Scheduling System for Nano-Materials Design and Simulation

Abstract

Available task scheduling systems can not support MPI parallel computing applications to be suspended for quickly inserting the emergency parallel computing tasks. By modifying TCP/IP protocol, this paper proposes a new method to solve the processes’ communication synchronization for suspending parallel application; moreover, by modifying the signal mechanism of the Linux operating system, this paper also proposes a method to solve the problems of consistently suspending and recovering parallel application. A Parallel computing dynamic task scheduling prototype system is implemented, and the experiment results show that the prototype system can suspend running parallel computing application, and also support dynamic insertion of emergency MPI parallel computing application.